Adapt gain as function of followup error, model can be used